MARL Spectrum Sharing in Vehicular Networks

To train the single-agent RL benchmark model, use the SARL training script with the environment and replay memory modules.

Required files:

• main_sarl_train.py
• Environment_marl.py
• replay_memory.py

# Run the SARL training script
python main_sarl_train.py

To evaluate trained models in a testing environment, use the testing suite. This setup allows you to reproduce Figures 3 and 4 from the original paper.

Requirements:

• Files: main_test.py, Environment_marl_test.py, replay_memory.py
• Model directory: A folder named /model containing the trained weights.

Customization: To change the V2V payload size, modify the self.demand_size attribute within Environment_marl_test.py.

Note: Avoid using the "Test" mode within main_marl_train.py as it is not recommended.

# Run the testing script
python main_test.py

The repository provides specific paths to reproduce results from the paper "Spectrum sharing in vehicular networks based on multi-agent reinforcement learning":

• Figures 3 and 4: Run main_test.py. You can adjust the V2V payload size by modifying self.demand_size in Environment_marl_test.py.
• Figure 5: This figure must be generated by recording the returns (performance metrics) during the training phase.
• Figures 6-7: These figures represent the performance of arbitrary episodes (typically showing successful MARL transmission vs. failed random baselines) and can be observed during testing.

To train the MARL model, you need to run the training script alongside the environment and replay memory modules. Ensure you have the required dependencies (Python 3.6 and TensorFlow 1.12.0) installed.

Required files:

• main_marl_train.py
• Environment_marl.py
• replay_memory.py

# Run the MARL training script
python main_marl_train.py